Hello,

I just found this patch and corresponding abstraction which is a neuron layer for PD. Great, enjoy the power of AI!

[url=http://www.la-kitchen.fr/augmented_mapping.html ]http://www.la-kitchen.fr/augmented_mapping.html

See the corresponding papers for details ...

Hi,

Thanks for posting the information about the neural network. I suspect that the biggest challenge for users will be implementing an appropriate training algorithm (e.g. like the backpropagation algorithm mentioned in Paul Churchland's The Engine of Reason, the Seat of the Soul). I wondered if any forum members had any suggestions or ideas regarding this.

I am not deeply familiar with neural network, I do understand the principle but I never experienced such DSP my self. My most complex application of "training" is rather simple adaptive DSP : LMS-gradient, leaky adaptation, and some others mostly related to adaptive DSP with linear time-invariant assumed filters, mostly transverse FIR.

So, I know that the author of the paper which I read from la kitchen was at this time using Matlab for the training. In the conclusion, they say that they work and training within PD.

My sole personnal reference on IA is some part of book like : "Signal Processing for Active Control", "Active control of noise and vibration" (from Hansen) and maybay Widrow "Adaptive Signal Processing" or "Adaptive Inverse Control" ...

Even outside PD, training its the biggest challenge ...

The linked PDF files on the hyperlink in the first post of this of this thread are not working for me. I am interested in the articles, but the files seem to errors. Any help?

The hyperlink still work for me ...

Here is the general info page :
[url=http://www.la-kitchen.fr/augmented_mapping.html ]http://www.la-kitchen.fr/augmented_mapping.html

Papers listening :
"Augmented Mapping: Towards an intelligent user-defined gesture mapping"
Arshia Cont, Thierry Coduys and Cyrille Henry.
Sound and Music Computing 2004 (SMC04), IRCAM - Centre George Pompidou, Paris, France.

The file :
[url=http://filebox.vt.edu/users/acont/html/SMC_Article.pdf ]http://filebox.vt.edu/users/acont/html/SMC_Article.pdf

"Real-time gesture mapping in Pd environment using Neural Networks"
Arshia Cont, Thierry Coduys and Cyrille Henry.
New Interfaces for Musical Expressions 2004 (NIME04), SUAC, Hamamatsu, Japan.

The file :
[url=http://filebox.vt.edu/users/acont/html/LaKitchen_NIME04.pdf ]http://filebox.vt.edu/users/acont/html/LaKitchen_NIME04.pdf

And some pd abstraction :
[url=http://www.la-kitchen.fr/NNgesture.zip ]http://www.la-kitchen.fr/NNgesture.zip

Everything is working fine for my part, even the pdf files.

thanks,
the links and files seem to work when i link to the from this forum page, but not off the la-kitchen page.
-zac

Thanks for the suggestions regarding training. I'll try to follow them up.

Best wishes,

David

Hello everyone,

I accidentally fell on this page and realized there are people interested in our work in La Kitchen. Let me know if you have any questions!

Regarding training, we are developing an external which does all the training for you.. . Stay tuned and it should be up soon! We have tested it and are thinking of a usable interface.. .

Cheers,

Arshia

The prospect of an external for training is very exciting, Arshia.

I will certainly try to follow developments here.

Best wishes,

David Roden

This is a pretty old topic but I would like to know if there is an implementation of neural network usable in pd.

I would like to build a midi effect that will feed a NN with midi notes, and velocities generated by random sequencer or a human player and then being able to activate the NN to read the patterns it has learnt as a sequencer.

Could anyone point to me any directions (patchs, articles) on how to do that ?

I actually found some articles, stuffs for Max Msp, I also discovered the ANN external for pd which should do the trick.

I just have to make this work now, if someone has already experimented with this I could use the help !